We’ve been providing hosting services to New Zealanders since 2004. The shape of our services has changed drastically in that time (just like the shape of our team, our office, and our data centre).
One thing that has stayed constant: we’ve always been on the look out for new ways to improve and expand the services we offer.
That was the thinking behind SiteHost’s freshly launched AI Platform. We’re hosting open-weight models here in Auckland, available to anyone with a SiteHost account. It’s a new way to offload automation and summarisation tasks securely and cost-effectively, with complete data privacy.
At SiteHost, we’re a team of developers always on the lookout for innovative tools to improve the way we work. We’re in the business of offering features that we know are really useful. That’s exactly the case for our AI Platform, which we’re already using in-house to handle lots of our transcription workload.
We’re excited to see how you’ll make the most of the AI models in your own creative and effective ways.
Carrying the weight
The progress open-weight AI models have made in recent years has increasingly impressed us. It’s at a point where they’re catching up with (and in some cases, surpassing) closed-source models.
Our appreciation of open-weight models is rooted in our long-standing preference for open-source technology. There’s no risk of vendor lock-in, and we’re not tethered to the commercial whims of a multinational enterprise.
We’re also cognisant that running models on your own hardware can be prohibitively expensive (especially with the ongoing hardware shortages and price spikes). For many businesses or individuals, the outlay needed for that kind of setup is tough to justify.
Bridging the gap
We’re making powerful open-weight models accessible with affordable pricing on our new AI Platform.
You can run open-weight AI models on our hardware, without any upfront costs or recurring monthly hosting expenses. Our AI Platform uses token-based charging. Depending on your situation, paying per request can be a lot more cost-effective than paying ongoing GPU hosting fees (it’s certainly cheaper upfront than forking out for your own dedicated hardware setup).
While premium monthly subscription-based services like Claude certainly have their place, usage limit restrictions make them inefficient for handling sporadic large data workloads.
And if you’re considering paying per-token for Claude or ChatGPT, it’s probably worth double-checking. The cheapest Claude model costs ~5x more per base input token than the most powerful model on our AI Platform. Even if the Claude models are more powerful per token, the cost difference is massive.
Provider | Model | Input / MTok | Cached input / MTok | Output / MTok |
SiteHost | Qwen/Qwen3-VL-Embedding-8B | NZ$0.09 | NZ$0.002 | — |
SiteHost | Qwen/Qwen3-VL-Reranker-8B | NZ$0.11 | NZ$0.004 | — |
SiteHost | Qwen/Qwen3.6-27B | NZ$0.28 | NZ$0.04 | NZ$2.76 |
OpenAI | gpt-4o-transcribe | US$2.50 | — | US$10.00 |
OpenAI | gpt-5.6.sol (Short context) | US$5.00 | US$0.50 | US$30.00 |
Anthropic | Claude Haiku 4.5 | US$1.00 | US$0.10 | US$5.00 |
Anthropic | Claude Opus 5 | US$5.00 | US$0.50 | US$25.00 |
Gemini 3.1 Flash TTS Preview | US$1.00 | — | US$20.00 | |
Gemini 3.6 Flash | US$1.50 | — | US$7.50 |
Prices as of 04 August 2026.
If you’re going to have predictably high-volume workloads or want to run custom proprietary models not offered by the AI Platform, then GPU Hosting could be worth considering. If your workload is large but more irregular, case-by-case, or demand-based, that’s where the models on the AI Platform shine.
Your data is your data
Lots of big multi-national AI providers have murky positions on data training policies.
Our AI Platform takes the opposite approach: we’re never storing any of your request data.
It's all hosted on servers in our data centre in Auckland, and the data never goes any further—the open-weight models are contained to our infrastructure and don’t ever send any data elsewhere.
New service, same goodness
Now you can run AI models and get all the SiteHost goodness you're already used to; local infrastructure in NZ, and one of the best support teams in the industry.
There’s tons of experience across our team to call upon, with our 24/7 support always available when you need it. We pride ourselves in being a trustworthy place to host important data and workloads, and it shows through in everything we do, from our infrastructure to our technical support. Hosting your automation workloads with us is no different.
Offloading the right work
Different AI models have their distinct strengths. When it comes to choosing a model, it’s mostly a balance of cost and scale efficiency.
Claude’s models are great tools for code generation and reasoning, but open-weight models like Qwen offer incredible performance for heavy data processing at a fraction of the cost. The Qwen models currently offered on our AI platform are less tailored toward generating code, and more about offloading processing and automation work. These models are optimised for workloads like text summarisation and data processing, and geared towards speeding up your work as cost effectively as possible.
Open-weight models have made AI technology more widely available and mouldable to each user’s unique needs. Now we’re deploying our top-of-the-line hardware to let customers make open-weight models work how you want.
Got data?
The models AI Platform offers are particularly useful at handling big blocks of difficult to understand data. If you have an archive of data in urgent need of tidying, sorting or summarising, then open-weight models are great at handling that.
While there’s all sorts of ways to take advantage of the powerful models, they’re generally best suited for data summarisation and transformation work. For example, we’re using the AI Platform in the office to handle our phone call transcription workload.
Alternatively, the embedding model we’re providing is useful for semantic searching through databases. You could feed it a search query and have it sift through a stack of documents for relevant passages.
However you see it fit, the models on our commercial-grade hardware are great at handling work with lots of data.
Plug & play
The AI Platform’s API endpoints mirror OpenAI’s formatting, so if you know how to use the OpenAI API then using it should feel familiar. You simply need to plug in your API key and you’ll be good to go, and can plug many other existing harnesses into it. You can also use our PHP & Python client libraries to run the service.
At this point we’re currently running three different Qwen models; one for inference work, one for embedding, and one specialising in reranking. We may add more in the future when new models emerge that fit the right use cases and performance/cost balance. We’ve also laid out retention and depreciation windows for each model, so you know you won’t be left in the lurch.
You can find a thorough overview of the AI Platform, and how it works, in our Knowledge Base.
The models can be used in many ways that’ll be unique to the organisation using them, so we’d love to hear any stories about the kinds of work you’re running on the AI Platform.
New possibilities
Hosting isn’t just about getting the fastest servers running your websites anymore. That’ll always be incredibly important, but the ways you can host data and websites is continuing to change. At SiteHost we’re always on the look out for new ways to improve and expand the services we can offer, and the AI Platform is the latest evidence of that.
Start by taking a look at our AI Platform or our Knowledge Base, or if there’s anything you want to know before jumping in just ask us.
Pricing information obtained on 04 July 2026 from:
https://platform.claude.com/docs/en/about-claude/pricing
https://developers.openai.com/api/docs/pricing?latest-pricing=standard
https://ai.google.dev/gemini-api/docs/pricing
https://cp.sitehost.nz/ai-platform/models